betaFeedback

WorksPhysics and Astronomy

Odd viscosity suppresses intermittency in direct turbulent cascades

Abstract

Intermittency refers to the broken self-similarity of turbulent flows caused by anomalous spatio-temporal fluctuations. In this Letter, we ask how intermittency is affected by a non-dissipative viscosity, known as odd viscosity (also Hall or gyro-viscosity), which appears in parity-breaking fluids such as magnetized polyatomic gases, electron fluids under magnetic field and spinning colloids or grains. Using a combination of Navier-Stokes simulations and theory, we show that intermittency is suppressed by odd viscosity at small scales. This effect is caused by parity-breaking waves, induced by odd viscosity, that break the multiple scale invariances of the Navier-Stokes equations. Building on this insight, we construct a two-channel helical shell model that reproduces the basic phenomenology of turbulent odd-viscous fluids including the suppression of anomalous scaling. Our findings illustrate how a fully developed direct cascade that is entirely self-similar can emerge below a tunable length scale, paving the way for designing turbulent flows with adjustable levels of intermittency.

In 3D turbulent fluids, energy is transferred from large to small length scales through a process known as a turbulent cascade: large eddies tend to split into smaller and smaller eddies [1, 2, 3, 4, 5, 6]. The hierarchical structure of the cascade suggests that it may exhibit a self-similarity relating larger and smaller eddies [7, 8]. It turns out that this self-similarity is only approximate [5]: smaller eddies exhibit an increasing propensity for extreme velocities, characterized by non-gaussian, fat-tailed distributions [9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29] (see Figure 1a). This phenomenon is known as intermittency [5, 11].

Image not extracted

Figure 1. Intermittency in odd turbulence. (a) Probability distribution of xx-direction vorticity in DNS (renormalized by the standard deviation), for νodd=0,3×10−4,6×10−4,1.2×10−3,2.4×10−3\nu_{\mathrm{odd}} = 0, 3 \times10^{-4}, 6 \times10^{-4}, 1.2 \times10^{-3}, 2.4 \times10^{-3}. Gaussian distribution is shown in dashed line for comparison. Inset: illustration of a chiral active fluid. Odd viscosity can be caused by self-spinning of particles in zz direction. (b) The growth of kurtosis KK in DNS is suppressed by odd viscosity at small scales, for same νodd\nu_{\mathrm{odd}} as in (a). Errorbar is shown as shaded area. (inset) Rescaled kurtosis K/K0K/K_0 (K0K_0 being the kurtosis for νodd=0\nu_{\mathrm{odd}} = 0) vs rescaled lengthscale koddr/πk_{\mathrm{odd}}r/\pi, with kodd=ϵodd−3/4k_{\mathrm{odd}} = \epsilon_{\mathrm{odd}}^{-3/4}. (c) Kurtosis predicted from the modified Parisi-Frisch theory, for νodd=0,10−6,10−5,10−4,10−3,10−2\nu_{\mathrm{odd}} = 0, 10^{-6}, 10^{-5}, 10^{-4}, 10^{-3}, 10^{-2}. (inset) Rescaled kurtosis K/K0K/K_0 vs rescaled lengthscale koddr/πk_{\mathrm{odd}}r/\pi. Note that the scale separation in (c) is considerably larger than what is achievable in DNS in (b). See Sec. II D for results with same scale separation as that in DNS (SI [60]).

The precise conditions that lead a turbulent cascade to produce intermittency are still unknown [1, 30]. In the direct cascade of the 3D Navier-Stokes equation, intermittency (i.e. non-gaussianity) increases along the forward cascade, a property known as “anomalous scaling”. Reductions of this tendency have been reported in magneto-hydrodynamics [31], nonlinear viscosity fluids [32], fluids on fractal Fourier sets [33–36] or subject to rotation [37–42]. It has been conjectured that self-similarity may be restored in a direct cascade by a mathematical “surgery” of Navier-Stokes equations that reduces the backward energy transfer through certain triadic interactions classified by the helicity of the modes involved [43]. However, we are not aware of any example of a fully developed direct cascade with no intermittency described by the 3D Navier-Stokes equations.

In this Letter, we show that the increase of intermittency, that typically occurs along a forward turbulent cascade, can be suppressed in the inertial range by parity-breaking waves occurring in fluids that display a non-dissipative viscosity, known as odd viscosity, such as magnetized polyatomic gases, graphene under magnetic field and spinning colloids or grains [44–59]. By controlling odd viscosity, we can tune the length scale below which the growth of intermittency terminates, ultimately creating a fully developed forward turbulent cascade that is completely self-similar, i.e. anomalous scaling is suppressed.

Turbulence with odd viscosity

While the viscosity tensor ηijkl\eta_{ijkl} is typically assumed to be symmetric (ηijkl=ηklij\eta_{ijkl} = \eta_{klij}), an antisymmetric part of the viscosity tensor (ηijklA=−ηklijA\eta_{ijkl}^{A} = -\eta_{klij}^{A}), known as odd, Hall or gyro viscosity, naturally arises in fluids that break both parity and time-reversal symmetry [56]. Odd viscosity with rotational symmetry in e^z\hat{e}_z (e.g., generated by self-spinning particles, see Figure 1(a)) modifies the incompressible Navier-Stokes equation to (see Supplementary Information (SI) [60])

Dtu=−∇P+νΔu+νodde^z×Δu+f(1)D_t \boldsymbol{u} = -\nabla P + \nu\Delta\boldsymbol{u} + \nu_{\mathrm{odd}}\hat{e}_z \times\Delta\boldsymbol{u} + \boldsymbol{f} \tag*{(1)}

where u(x,t)\boldsymbol{u}(x,t) is the velocity, Dt=∂t+u⋅∇D_t = \partial_t+\boldsymbol{u}\cdot\nabla the convective derivative, f\boldsymbol{f} the helicity-free external forcing, ν\nu the dissipative viscosity, e^z\hat{e}_z the unit vector along zz, and νodd\nu_{\mathrm{odd}} the odd viscosity. The odd viscosity term, that can be seen as a scale-dependent Coriolis force, does not directly dissipate or transfer energy, but affects energy transfer by inducing “odd waves” with frequency ω±(k)=±νoddkz∣k∣\omega_{\pm}(k) = \pm\nu_{\mathrm{odd}}k_z\lvert k\rvert in the fluid (SI) [60]. Odd waves are parity-breaking, i.e. the frequency changes sign when the definite helicity ±\pm is flipped [57]. As shown in Ref. [61], odd waves progressively arrest the forward energy flux starting from the characteristic scale kodd=ϵ1/4νodd−3/4k_{\mathrm{odd}} = \epsilon^{1/4}\nu_{\mathrm{odd}}^{-3/4} (ϵ\epsilon is the energy injection rate) where ω±(k)\omega_{\pm}(k) is comparable with the eddy turnover rate. While a qualitative picture of the energy spectrum of odd turbulence has emerged [61], the effect of odd viscosity on intermittency is yet to be elucidated.

Intermittency in odd turbulence—We first analyze the vorticity distribution obtained from direct numerical simulations (DNS), which captures the intermittency around the dissipation scale. Details of the DNS method are provided in SI [60]. Remarkably, odd viscosity leads to more Gaussian-like distributions (compare the different curves in Figure 1a). Intermittency can be captured by the kurtosis K=S4/S22K = S_4/S_2^2 of the structure functions Sp(r)=⟨(δru)p⟩S_p(r) = \langle(\delta_r u)^p\rangle in which δru=[u(x+r)−u(x)]⋅r^\delta_r u = [u(x + r) - u(x)] \cdot\hat{r} are known as the longitudinal velocity increment. Due to the anisotropy induced by odd viscosity, the structure function is evaluated for r^\hat{r} perpendicular to e^z\hat{e}_z. At the injection scale, the flow has a Gaussian kurtosis K=3K = 3 imposed by the forcing. In ordinary turbulence (νodd=0\nu_{\mathrm{odd}} = 0), the kurtosis increases as energy cascades towards smaller scales (Figure 1b). For finite νodd\nu_{\mathrm{odd}}, the growth of the kurtosis is suppressed at small scales where odd viscosity is dominant. The rescaled data (inset) shows that the transition between the intermittent and non-intermittent regimes is located around r=π/koddr = \pi/k_{\mathrm{odd}}, where odd viscosity becomes relevant [61].

Modified Parisi-Frisch formalism—To rationalize this suppression of intermittency, we revisit the Parisi-Frisch framework [62] relating intermittency to the existence of multiple scale invariances in Navier-Stokes equations.

In the absence of odd viscosity, the unforced dissipation-free Navier-Stokes equation is invariant under the scale transformation x→λxx \to\lambda x, u→λhuu \to\lambda^h u, t→λ1−htt \to\lambda^{1-h}t for all hh. Kolmogorov’s K41 theory [7, 8] effectively assumes that the exponent h=1/3h = 1/3 is selected, leading to a self-similar solution. In reality, turbulent flows exhibit a statistical distribution Pr(h)∼rF(h)P_r(h) \sim r^{F(h)} of scaling exponents hh, as allowed by the scaling symmetry. This intertwining of different exponents hh, referred to as multifractality, breaks the self-similarity of the flow statistics at different scales, heralding the emergence of intermittency [5] [63].

Intermittency is suppressed by odd viscosity because it violates the multiple scaling symmetry of Navier-Stokes equations (SI [60]). We conjecture that intermittent solutions δru∼rh\delta_r u \sim r^h with arbitrary hh can still appear at large scales where νodd\nu_{\mathrm{odd}} has a small effect, and that they reduce

to the same wave-affected solution δru∼νodd1/4\delta_r u\sim\nu_{\mathrm{odd}}^{1/4} at small scales, governed by a single exponent h=0h=0 (SI [60]).

We model this crossover by adapting the approach of Ref. [65] and obtain

δru=rh[1+rodd(h)/r]F(h),Pr(h)=rF(h)[1+rodd(h)/r]F(h).(2)\begin{aligned} \delta_r u &= r^h\left[1+r_{\mathrm{odd}}(h)/r\right]^{F(h)},\\ P_r(h) &= r^{F(h)}\left[1+r_{\mathrm{odd}}(h)/r\right]^{F(h)}. \tag*{(2)} \end{aligned}

in which rodd(h)=νodd1/(1+h)r_{\mathrm{odd}}(h)=\nu_{\mathrm{odd}}^{1/(1+h)} is the characteristic lengthscale, which is determined by equating the eddy turnover time τeddy=r/δru\tau_{\mathrm{eddy}}=r/\delta_r u and the odd wave period τodd=νodd−1r2\tau_{\mathrm{odd}}=\nu_{\mathrm{odd}}^{-1}r^2 [61]. Eq. (2) qualitatively reproduces the arrest of intermittency observed in DNS, see Figure 1c (F(h)F(h) is adopted from Ref. [63]). For small odd viscosity, K(r)K(r) is non-monotonic, in agreement with DNS with extended inertial range (SI [60]). This is due to the variation of rodd(h)r_{\mathrm{odd}}(h), such that the crossover between two regimes takes place at different lengthscale for different hh. Similar non-monotonicity was observed in Ref. [65].

Nonlinear energy transfer—In the Navier-Stokes equa-

tion, nonlinear energy transfer occurs through triads of modes with wavenumbers satisfying k+p+q=0\mathbf{k}+\mathbf{p}+\mathbf{q}=0 [30] [66, 67, 68, 69]. As helicity is conserved, it is convenient to consider modes with definite helicity sk=±1s_k=\pm1. When odd waves are present, these modes get out of phase. As a consequence, the contribution to the non-linear energy transfer of a given triad is multiplied by exp⁡(iω~t)\exp(i\widetilde{\omega}t) in which ω~=ωsk(k)+ωsp(p)+ωsq(q)\widetilde{\omega}=\omega_{s_k}(k)+\omega_{s_p}(p)+\omega_{s_q}(q) [61]. When ω~τeddy≫1\widetilde{\omega}\tau_{\mathrm{eddy}}\gg1, the fast oscillation effectively cancels the time-averaged energy transfer during τeddy\tau_{\mathrm{eddy}}. Therefore, in odd turbulence energy can only transfer through quasi-resonant triads (for which ω~τeddy≤1\widetilde{\omega}\tau_{\mathrm{eddy}}\leq1) [70]. We estimate the average energy transfer from the fraction gkg_k of quasi-resonant triads and the energy transfer rate ϵ∼kuk3\epsilon\sim ku_k^3 in ordinary turbulence, leading to ϵ∼gkuk3\epsilon\sim g_ku_k^3.

For quasi local triads (k=∣k∣≈∣p∣≈∣q∣k=|\mathbf{k}|\approx|\mathbf{p}|\approx| \mathbf{q}|) [71] we have ω~∼νoddk2\widetilde{\omega}\sim\nu_{\mathrm{odd}}k^2 and τeddy=k−1uk−1\tau_{\mathrm{eddy}}=k^{-1}u_k^{-1}, hence gk=1g_k=1 for νodd≪uk/k\nu_{\mathrm{odd}}\ll u_k/k, corresponding to a regular turbulence regime for k≪koddk\ll k_{\mathrm{odd}}. For νodd≫uk/k\nu_{\mathrm{odd}}\gg u_k/k (k≫koddk\gg k_{\mathrm{odd}}), odd waves arrest the energy transfer with gk∼uk/(kνodd)g_k\sim u_k/(k\nu_{\mathrm{odd}}) and ϵ∼uk4/νodd\epsilon\sim u_k^4/\nu_{\mathrm{odd}}. Interestingly, odd waves have a strong impact on the forward transfer of heterochiral triads (+−−+,−+−)(+--+, -+-) but only a weak impact on the inverse transfer of homochiral triads (+++,−−−)(+++,---) [43, 72], see Figure 2(a), because for homochiral triads with exact locality we always have ωˉ=0\bar{\omega}=0 and gk=1g_k=1 [73]. This asymmetry enhances the flux loop formed by the two channels (Figure 3b) and triggers the aforementioned breakdown of the multiple scale invariances of Navier-Stokes equations, which ultimately suppresses intermittency (SI [60]). Importantly, such a mechanism requires the energy to be symmetrically distributed on modes with both definite helicities (+ and −), which ensures that the energy flux is not dominated by homochiral channels only.

Image not extracted

Figure 2. Effect of odd waves on triadic interactions and helical shell model. (a) Asymmetric effects of odd viscosity on homochiral and heterochiral triads. Odd viscosity strongly affects heterochiral triads by reducing the fraction of resonant triads, gkg_k. Odd viscosity has no effect on local homochiral triads for which gk=1g_k = 1. (b) Two-channel helical shell model. Each shell contains two velocity variables representing modes with ++ and −- chiralities, respectively. Energy transfers forward in heterochiral channel and backward in homochiral channel, through triads formed by nearest neighbors.

Image not extracted

Figure 3. Odd turbulence in DNS and the helical shell model. (a). Energy spectrum E(k)E(k) in DNS, for νodd=0\nu_{\mathrm{odd}}=0, 3×10−43\times10^{-4}, 6×10−46\times10^{-4}, 1.2×10−31.2\times10^{-3}, 2.4×10−32.4\times10^{-3}. The spectrum in the inertial range changes from a K41 regime (E(k)∼k−5/3E(k)\sim k^{-5/3}) to a wave-affected regime (E(k)∼k−1E(k)\sim k^{-1}). (inset) Rescaled spectrum E(k)/E0E(k)/E_0 (E0E_0 being the spectrum for νodd=0\nu_{\mathrm{odd}}=0) vs rescaled wave number k/koddk/k_{\mathrm{odd}}, with kodd=ϵodd−3/4k_{\mathrm{odd}}=\epsilon_{\mathrm{odd}}^{-3/4}. (b) Energy flux Π(k)\Pi(k) and flux loop formed by homochiral and heterochiral components in DNS. νodd=2.4×10−3\nu_{\mathrm{odd}}=2.4\times10^{-3}. (c) Kurtosis in DNS, same as Figure 1(d) with x-axis changed from rr to π/r\pi/r. (inset) Rescaled kurtosis K/K0K/K_0 (K0K_0 being the kurtosis for νodd=0\nu_{\mathrm{odd}}=0) vs rescaled lengthscale π/(koddr)\pi/(k_{\mathrm{odd}}r). (d) Energy spectrum in the shell model, for νodd=0\nu_{\mathrm{odd}}=0, 2.6×10−72.6\times10^{-7}, 1.0×10−61.0\times10^{-6}, 4.1×10−64.1\times10^{-6}, 1.6×10−51.6\times10^{-5}, 6.6×10−56.6\times10^{-5}. (inset) Rescaled spectrum E(k)/E0E(k)/E_0 vs rescaled wave number k/koddk/k_{\mathrm{odd}}. (e) Energy fluxes in the shell model, for νodd=6.6×10−5\nu_{\mathrm{odd}}=6.6\times10^{-5}. (f) Kurtosis K(kn)=S4(kn)/S22(kn)K(k_n)=S_4(k_n)/S_2^2(k_n) truncated at the dissipation scale, with Sp(kn)=⟨(∣un+∣∣un+1+∣)p/2⟩S_p(k_n)=\langle(|u_n^+||u_{n+1}^+|)^{p/2}\rangle [29] [64], for νodd\nu_{\mathrm{odd}} values same as that in (d). (inset) Rescaled kurtosis K/K0K/K_0 vs rescaled wavenumber k/koddk/k_{\mathrm{odd}}. In the shell model simulations, we have used N=30N=30 shells, λ=2\lambda=2, k0=1/8k_0=1/8, an injection rate ϵ=10−5\epsilon=10^{-5}, β=0.3\beta=0.3, γ=0.2\gamma=0.2 and ν=10−9\nu=10^{-9}. A constant injection forcing fn±=0.5δn,3ϵ/un±∗f_n^\pm=0.5\delta_{n,3}\epsilon/u_n^{\pm*} is used. Note that the scale separation in the shell model (bottom row) is considerably larger than what is achievable in DNS (top row). See Sec. II D for results with same scale separation as that in DNS (SI [60]).

Helical shell model

We now construct a two-channel helical shell model that captures most odd-turbulence phenomenology (Figure 2b) [74]. Shell models have proven to be a particularly useful tool to study the inertial-range intermittency in turbulence, as they can probe scales far beyond what is accessible in DNS [19] (Although they behave differently from DNS at the dissipation scale [65]). Shells are equispaced on a logarithmic scale kn=k0λnk_n=k_0\lambda^n and each shell has two complex dynamical variables, un+u_n^+ and un−u_n^-, representing velocity fluctuations of modes with + and − definite helicities, respectively. We impose conservation of energy E=∑n∣un+∣2+∑n∣un−∣2E=\sum_n|u_n^+|^2+\sum_n|u_n^-|^2 and helicity H=∑nkn∣un+∣2−∑nkn∣un−∣2H=\sum_n k_n|u_n^+|^2-\sum_n k_n|u_n^-|^2 [76], resulting in

∂tun±=Bn±,∓+Bn±,±+fn±−νkn2un±,(3)\partial_tu_n^\pm=B_n^{\pm,\mp}+B_n^{\pm,\pm}+f_n^\pm-\nu k_n^2u_n^\pm, \tag*{(3)}

where fnf_n is the forcing, and

Bn+,→+=an+,→+un+2+un+1+∗+bn+,→+un+1+un−1+∗−cn+,→+un−1+un−2+Bn+,→−=an+,→−un+2+un+1−∗+bn+,→−un+1+un−1−∗−cn+,→−un−1+un−2−∗\begin{aligned} B_n^{+,\to+}&=a_n^{+,\to+}u_{n+2}^+u_{n+1}^{+*}+b_n^{+,\to+}u_{n+1}^+u_{n-1}^{+*}-c_n^{+,\to+}u_{n-1}^+u_{n-2}^+\\ B_n^{+,\to-}&=a_n^{+,\to-}u_{n+2}^+u_{n+1}^{-*}+b_n^{+,\to-}u_{n+1}^+u_{n-1}^{-*}-c_n^{+,\to-}u_{n-1}^+u_{n-2}^{-*} \end{aligned}

are the nonlinear transfer of heterochiral and homochiral channels, illustrated in purple and green in Figure 2. When an+→+=an+→−=kna_n^{+\to+}=a_n^{+\to-}=k_n, the channels reduce to the Sabra shell models for the forward and inverse cascade regimes, respectively [64, 77–80] (bnb_n and cnc_n are determined from conservation laws, SI [60]). To reproduce the effect of odd viscosity, we let

an+→∓=kngn+→∓=kn(1+νoddkn∣un+2∓∣+∣un+1∓∣+∣un∓∣)−1a_n^{+\to\mp}=k_ng_n^{+\to\mp}=k_n\left(1+\frac{\nu_{\mathrm{odd}}k_n}{|u_{n+2}^{\mp}|+|u_{n+1}^{\mp}|+|u_n^\mp|}\right)^{-1}
an+→∓=βkngn+→∓=βkn(1+γνoddkn∣un+2±∣+∣un+1±∣+∣un±∣)−1a_n^{+\to\mp}=\beta k_ng_n^{+\to\mp}=\beta k_n\left(1+\frac{\gamma\nu_{\mathrm{odd}}k_n}{|u_{n+2}^{\pm}|+|u_{n+1}^{\pm}|+|u_n^\pm|}\right)^{-1}

where 0<β<10<\beta<1 ensures that the model is dominated by the forward cascade when νodd=0\nu_{\mathrm{odd}}=0. This functional form is chosen to introduce a transition around νodd∼∣un∣/kn\nu_{\mathrm{odd}}\sim|u_n|/k_n. Here, gng_n mimics the fraction of resonant triads: gn≈1g_n\approx1 for small νodd\nu_{\mathrm{odd}} and gn∼∣un∣/(knνodd)g_n\sim|u_n|/(k_n\nu_{\mathrm{odd}}) for large νodd\nu_{\mathrm{odd}}. Finally, we choose 0<γ<10<\gamma<1 so that the suppression effect is weaker for the homochiral channel. In the SI, we generalize the model to include “generalized odd viscosity”, for which ω±(k)=∓coddkz∣k∣α−1\omega_\pm(k)=\mp c_{\mathrm{odd}}k_z|k|^{\alpha-1} [60].

As shown in Figure 3e, results from the numerical simulation of this shell model reproduce the energy flux observed in Navier-Stokes simulation for k>koddk>k_{\mathrm{odd}} as well as other qualitative features and scaling relations (Figure 3a, Figure 3d). In the SI [60], we show that a flux loop state [30] consistent with DNS results [61] is observed in the inverse cascade regime where forcing is injected at small lengthscale kin>koddk_{\mathrm{in}}>k_{\mathrm{odd}}.

Suppression of anomalous scaling

Our shell model results (Figure 3f) mimic the DNS kurtosis (Figure 3c) and its non-monotonic dependence predicted by the modified Parisi-Frisch argument. We also verify that intermittency is suppressed in higher-order structure functions up to p=8p=8, in both DNS and shell model (SI [60]). For generalized odd viscosity with α>2/3\alpha>2/3, we find a wave-affected scaling regime of the structure functions for k≫koddk\gg k_{\mathrm{odd}}, where the wave-affected exponents ζp\zeta_p are extracted (see Figure 4(a) for an example). Figure 4(b) shows that the ζp\zeta_p extracted in the wave-affected regime are consistent with the self-similar exponents ζp=hαp\zeta_p=h_\alpha p (solid lines), in contrast with the anomalous exponents usually observed in forward cascades (black dashed line). The hαh_\alpha we measure are close to the theoretical approximation hα≈1/2−α/4h_\alpha\approx1/2-\alpha/4 (SI [60]).

Image not extracted

Figure 4. Scaling exponents in the helical shell model. (a) Rescaled structure function kodd4/3S4k_{\mathrm{odd}}^{4/3}S_4 (kodd=ϵ1/(3α−2)codd3/(2−3α)k_{\mathrm{odd}}=\epsilon^{1/(3\alpha-2)}c_{\mathrm{odd}}^{3/(2-3\alpha)}) in shell model with generalized odd viscosity for codd/c0=0,4mc_{\mathrm{odd}}/c_0=0,4^m, with c0=(ϵ/2)1/2−α/4νodd3α/4−1/2c_0=(\epsilon/2)^{1/2-\alpha/4}\nu_{\mathrm{odd}}^{3\alpha/4-1/2}, α=1.5\alpha=1.5 and m=3−7m=3-7. (b) Exponents in the wave-affected regime ζp\zeta_p for α=1.2,1.5,2\alpha=1.2,1.5,2. Exponents for α=1.5\alpha=1.5 are fitted by the shaded area in (a), see SI for details [60]. Solid lines are monofractal fit ζp=hαp\zeta_p=h_\alpha p. Error bars are estimated from shifting the fitting range by ±1\pm1 shells. Exponents for codd=0c_{\mathrm{odd}}=0 are shown for comparison, which is consistent with She-Lévêque (dashed line) [63]. (inset) ζp/ζ3\zeta_p/\zeta_3 for codd=0c_{\mathrm{odd}}=0 and for α=1.5\alpha=1.5. Shell model parameters are same as that in Figure 3. Inertial range in the simulation is extended (SI [60]).

To sum up, we have demonstrated the existence of a fully developed forward turbulent energy cascade that is entirely self-similar below a tunable length scale, paving the way for designing turbulent flows with adjustable levels of intermittency. In future work it would be interesting to generalize the mechanism studied here to more complex flows where other sources of waves may exist [81].

Acknowledgments

We thank Alexandros Alexakis, Luca Biferale and Tali Khain for discussions. We are grateful for the support of the Netherlands Organisation for Scientific Research (NWO) for the use of supercomputer facilities (Snellius) under Grant No. 2021.035, 2023.026. This publication is part of the project “Shaping turbulence with smart particles” with project number OCENW.GROOT.2019.031 of the research programme Open Competitie ENW XL which is (partly) financed by the Dutch Research Council (NWO). S.C. and M.F. acknowledge a Kadanoff–Rice fellowship funded by the National Science Foundation under award no. DMR-2011854. M.F. acknowledges partial support from the National Science Foundation under grant DMR-2118415 and the Simons Foundation. V.V. acknowledges partial support from the Army Research Office under grant W911NF-22-2-0109 and W911NF-23-1-0212. M.F. and V.V acknowledge partial support from the France Chicago center through a FACCTS grant. This research was partly supported from the National Science Foundation through the Center for Living Systems (grant no. 2317138), the National Institute for Theory and Mathematics in Biology and the Chan Zuckerberg Foundation. This work was completed in part with resources provided by the University of Chicago’s Research Computing Center.

Data availability

The code used for simulating the helical shell model, processing the data, generating the figures, as well as the data generated during this study are available on Zenodo [82] under the 2-clause BSD licence.

Supplementary Information:

Odd viscosity suppresses intermittency in direct turbulent cascades

Odd viscosity

In this section we introduce odd viscosity and emphasize its non-dissipative nature.

In fluids, the stress tensor σij\sigma_{ij} is related to the strain rate via a viscosity tensor ηijkl\eta_{ijkl}: σij=ηijkl∂luk\sigma_{ij}=\eta_{ijkl}\partial_l u_k. The viscosity tensor is usually assumed to be symmetric, i.e., ηijkl=ηklij\eta_{ijkl}=\eta_{klij}, which applies to passive fluids. However, in active and parity-breaking fluids, e.g., fluids formed by self-spinning particles, asymmetric viscosity tensor (ηijkl≠ηklij\eta_{ijkl}\ne\eta_{klij}) arises because the system is out of equilibrium [1]. We decompose the viscosity tensor into a symmetric part ηijklS=(ηijkl+ηklij)/2\eta^{S}_{ijkl}=(\eta_{ijkl}+\eta_{klij})/2 and an antisymmetric part ηijklA=(ηijkl−ηklij)/2\eta^{A}_{ijkl}=(\eta_{ijkl}-\eta_{klij})/2. The incompressible Navier-Stokes equation reads:

ρ0Dtuj=−∇jP+∂i(ηijkl∂luk)=−∇jP+η∇uj+∂i(ηijklA∂luk),(4)\rho_0 D_t u_j=-\nabla_j P+\partial_i(\eta_{ijkl}\partial_l u_k) =-\nabla_j P+\eta\nabla u_j+\partial_i(\eta^{A}_{ijkl}\partial_l u_k), \tag*{(4)}

where ρ0\rho_0 is the density. In the last equality, the second term is the dissipation term that arises from ηijklS\eta^{S}_{ijkl}.

In this work we consider odd viscosity generated by the self-spinning of particles in e^z\hat{e}_z. This corresponds to a cylindrically symmetric viscosity tensor which has eight independent odd viscosities [2]. Only two independent odd viscosity terms η1o\eta^o_1 and η2o\eta^o_2 remain in the incompressible Navier-Stokes equation [3]:

ρ0Dtu=−∇P+ηΔu+η1o[(∂x2+∂y2)uy−(∂x2+∂y2)ux0]+η2o[−∂z2uy−∂y∂zuz∂z2ux+∂x∂zuz∂z(∂yux−∂xuy)](5)\rho_0 D_t\boldsymbol{u}=-\nabla P+\eta\Delta\boldsymbol{u} +\eta^o_1 \begin{bmatrix} (\partial_x^2+\partial_y^2)u_y\\ -(\partial_x^2+\partial_y^2)u_x\\ 0 \end{bmatrix} +\eta^o_2 \begin{bmatrix} -\partial_z^2u_y-\partial_y\partial_z u_z\\ \partial_z^2u_x+\partial_x\partial_z u_z\\ \partial_z(\partial_yu_x-\partial_xu_y) \end{bmatrix} \tag*{(5)}

One example of the corresponding odd viscosity tensor is

ηijklA=η1o2(δijδikϵil3+δijδilϵik3−δklδkiϵkj3−δklδkjϵki3)+η2o2(δj3δl3ϵki3+δj3δk3ϵli3+δi3δl3ϵkj3+δi3δk3ϵlj3)(6)\eta^A_{ijkl}=\frac{\eta^o_1}{2}\left(\delta_{ij}\delta_{ik}\epsilon_{il3}+\delta_{ij}\delta_{il}\epsilon_{ik3}-\delta_{kl}\delta_{ki}\epsilon_{kj3}-\delta_{kl}\delta_{kj}\epsilon_{ki3}\right)+\frac{\eta^o_2}{2}\left(\delta_{j3}\delta_{l3}\epsilon_{ki3}+\delta_{j3}\delta_{k3}\epsilon_{li3}+\delta_{i3}\delta_{l3}\epsilon_{kj3}+\delta_{i3}\delta_{k3}\epsilon_{lj3}\right) \tag*{(6)}

In Eq. (6) Einstein summation is not used. Note that Eq. (6) is not the only possible viscosity tensor that gives Eq. (5). In our DNS we assume η1o=−2η2o\eta^o_1=-2\eta^o_2 and η2o=ρ0νodd\eta^o_2=\rho_0\nu_{\mathrm{odd}}. In this case the force due to odd viscosity reduces to ρ0νodd[e^z×Δu+∇(∂yux−∂xuy)]\rho_0\nu_{\mathrm{odd}}[\hat{e}_z\times\Delta\boldsymbol{u}+\nabla(\partial_yu_x-\partial_xu_y)], where the gradient term can be absorbed into the pressure. Equation (5) then reduces to

Dtu=−∇P+νΔu+νodde^z×Δu+f(7)D_t\boldsymbol{u}=-\nabla P+\nu\Delta\boldsymbol{u}+\nu_{\mathrm{odd}}\hat{e}_z\times\Delta\boldsymbol{u}+\boldsymbol{f} \tag*{(7)}

which is Eq. (1) of the main text.

In Ref. [2], it is proved that the asymmetric viscosity tensor ηijklA\eta^A_{ijkl} does not generate or dissipate energy. For an intuitive understanding, we consider the following reduced dynamical equation of the odd viscosity term:

∂tu=νodde^z×Δu(8)\partial_t\boldsymbol{u}=\nu_{\mathrm{odd}}\hat{e}_z\times\Delta\boldsymbol{u} \tag*{(8)}

We perform Fourier transform u=∑ku~(k,t)eik⋅x\boldsymbol{u}=\sum_{\boldsymbol{k}}\tilde{\boldsymbol{u}}(\boldsymbol{k},t)e^{i\boldsymbol{k}\cdot\boldsymbol{x}}, which leads to

∂t[u~x(k,t)u~y(k,t)u~z(k,t)]=νoddk2[0−10100000][u~x(k,t)u~y(k,t)u~z(k,t)](9)\partial_t \begin{bmatrix} \tilde{u}_x(\boldsymbol{k},t)\\ \tilde{u}_y(\boldsymbol{k},t)\\ \tilde{u}_z(\boldsymbol{k},t) \end{bmatrix} =\nu_{\mathrm{odd}}k^2 \begin{bmatrix} 0 & -1 & 0\\ 1 & 0 & 0\\ 0 & 0 & 0 \end{bmatrix} \begin{bmatrix} \tilde{u}_x(\boldsymbol{k},t)\\ \tilde{u}_y(\boldsymbol{k},t)\\ \tilde{u}_z(\boldsymbol{k},t) \end{bmatrix} \tag*{(9)}

The antisymmetric, non-Hermitian dynamical matrix has imaginary eigenvalues ±iνoddk2\pm i\nu_{\mathrm{odd}}k^2, which is distinct from that of positive viscosity (negative eigenvalues) and negative viscosity (positive eigenvalues). Equation (9) has oscillatory solutions u(k,t)∼e±iνoddk2t\boldsymbol{u}(\boldsymbol{k},t)\sim e^{\pm i\nu_{\mathrm{odd}}k^2t}, i.e., energy is neither generated nor dissipated by odd viscosity. Note that in Eq. (8) the incompressibility ∇⋅u=0\nabla\cdot\mathbf{u} = 0 is not included. Imposing the incompressibility changes the eigenvalues to ±iνodd∣k∣kz\pm i\nu_{\mathrm{odd}}|\mathbf{k}|k_z [2, 3].

Above we have shown that odd viscosity is non-dissipative. As a linear viscosity term, it does not directly affect the nonlinear triadic interaction. Hence, the energy balance equation follows that of the ordinary turbulence [4, 5, 6]:

∂tE=−2νk2E−T+F,(10)\partial_t E = -2\nu k^2 E - T + F, \tag*{(10)}

in which

T(k,t)=Im⁡∑p+q=ku~i∗(k,t)Pij(k)qℓu~ℓ(p,t)u~j(q,t).(11)T(\mathbf{k},t) = \operatorname{Im}\sum_{\mathbf{p}+\mathbf{q}=\mathbf{k}} \widetilde{u}_i^*(\mathbf{k},t)P_{ij}(\mathbf{k})q_\ell\widetilde{u}_\ell(\mathbf{p},t)\widetilde{u}_j(\mathbf{q},t). \tag*{(11)}

This term describes the non-linear energy transfer between scales, while F=u∗⋅fF = \mathbf{u}^* \cdot\mathbf{f} corresponds to energy injection by the forcing term f\mathbf{f}. The term −2νk2E-2\nu k^2 E represents standard viscous dissipation. In Eq. (11), the sum runs on momenta p\mathbf{p} and q\mathbf{q} such that p+q=k\mathbf{p}+\mathbf{q}=\mathbf{k}, and Pij(k)=δij−kikj/k2P_{ij}(\mathbf{k}) = \delta_{ij} - k_i k_j/k^2 is the projector on incompressible flows [5, 6, 7, 8, 9].

Simulation Details

We perform direct numerical simulations (DNS) of the Navier-Stokes equation incorporating odd viscosity (Eq. (1) in the main text) within a cubic domain of size L=2πL = 2\pi, employing periodic boundary conditions. Our approach utilizes a pseudo-spectral method coupled with Adams-Bashforth time-stepping and a 2/3-dealiasing rule [10]. The dissipative and odd viscous terms are integrated exactly using integrating factors. The forcing f(t,k)\mathbf{f}(t,\mathbf{k}) acts within a wavenumber band k∈[kin,kin+1]k \in[k_{\mathrm{in}}, k_{\mathrm{in}}+1], with random phases delta-correlated in space and time, ensuring a consistent average energy injection rate ϵ=⟨u⋅f⟩\epsilon= \langle\mathbf{u}\cdot\mathbf{f}\rangle. This forcing possesses a zero mean component ⟨f(t,k)⟩=0\langle\mathbf{f}(t,\mathbf{k})\rangle= \mathbf{0} and covariance ⟨f(t,k)⋅f(t′,k′)⟩=ϵδ(t−t′)δ(k−k′)\langle\mathbf{f}(t,\mathbf{k})\cdot\mathbf{f}(t',\mathbf{k}')\rangle= \epsilon\delta(t-t')\delta(\mathbf{k}-\mathbf{k}'). The time-step is selected to resolve the fastest odd wave with frequency τodd,max−1=νoddkmax2\tau_{\mathrm{odd,max}}^{-1} = \nu_{\mathrm{odd}}k_{\mathrm{max}}^2, where kmaxk_{\mathrm{max}} denotes the highest resolved wavenumber in the domain. We observe that stable integration requires a time step Δt≲0.1τodd,max\Delta t \lesssim0.1\tau_{\mathrm{odd,max}}.

Since the forcing is helicity-free, and odd viscosity conserves helicity, the net total helicity of the flow remains identically zero.

The parameters used for the simulations in the main text are as in Ref. [3]. In Section II C we show additional simulations with increased resolution and hyperdissipation to further extend the inertial range. All simulation parameters are provided in Table I.

⟨ϵ⟩\langle\epsilon\rangleν\nuνodd\nu_{\mathrm{odd}}kink_{\mathrm{in}}koddk_{\mathrm{odd}}N3N^3ℓν/(L/N)\ell_\nu/(L/N)TTΔt\Delta tτν/Δt\tau_\nu/\Delta tRe\mathrm{Re}
I1.4×10−51.4\times10^{-5}9.4×10−69.4\times10^{-6}[0.3−2.4]×10−3[0.3-2.4]\times10^{-3}3[27−6][27-6]7683768^30.342×1032\times10^3[1.0−0.2]×10−2[1.0-0.2]\times10^{-2}[82−410][82-410]1.3×1031.3\times10^3
II1.0[2−20]×10−10[2-20]\times10^{-10} (*)2.0×10−22.0\times10^{-2}218153631536^30.350.65×10−55\times10^{-5}2002.5×1032.5\times10^3

Table I. Parameters that are used for DNS in the main text (I) and in Section II C (II). Listed are the average energy injection rate ⟨ϵ⟩\langle\epsilon\rangle, normal shear viscosity ν\nu, odd viscosity νodd\nu_{\mathrm{odd}}, injection wavenumber kink_{\mathrm{in}}, odd viscosity wavenumber kodd=ϵ−3/4νodd3/4k_{\mathrm{odd}}=\epsilon^{-3/4}\nu_{\mathrm{odd}}^{3/4}, grid resolution N3N^3, Kolmogorov length ℓν=ϵ−1/4ν3/4\ell_\nu=\epsilon^{-1/4}\nu^{3/4}, total simulation time TT, time-step Δt\Delta t, Kolmogorov time τν=ϵ−1/2ν1/2\tau_\nu=\epsilon^{-1/2}\nu^{1/2} and the a posterioria\ posteriori integral scale Reynolds number without odd viscosity Re\mathrm{Re}.

(*) hyperdissipation (quadratic Laplacian)

Shell Model

In the two-channel helical shell model, shells are logarithmically equispaced with kn=k0λnk_n = k_0\lambda^n. The dynamic equation reads (Eq. (3) of the main text)

∂tun±=Bn±,→+Bn±,←+fn±−νkn2un±,(12)\partial_t u_n^\pm= B_n^{\pm,\to} + B_n^{\pm,\leftarrow} + f_n^\pm- \nu k_n^2 u_n^\pm, \tag*{(12)}

where fnf_n is the forcing.

Bn+,→=an+,→un+2+un+1−∗+bn+,→un+1+un−1−∗−cn+,→un−1+un−2+,(13)B_n^{+,\to} = a_n^{+,\to}u_{n+2}^+u_{n+1}^{-*} + b_n^{+,\to}u_{n+1}^+u_{n-1}^{-*} - c_n^{+,\to}u_{n-1}^+u_{n-2}^{+}, \tag*{(13)}
Bn+,←=an+,←un+2+un+1−∗+bn+,←un+1+un−1−∗−cn+,←un−1+un−2+(14)B_n^{+,\leftarrow} = a_n^{+,\leftarrow}u_{n+2}^+u_{n+1}^{-*} + b_n^{+,\leftarrow}u_{n+1}^+u_{n-1}^{-*} - c_n^{+,\leftarrow}u_{n-1}^+u_{n-2}^{+} \tag*{(14)}

are the nonlinear transfer of heterochiral and homochiral channels.

an±,∓=kngn±,∓=kn(1+νoddkn∣un+2±∣+∣un+1∓∣+∣un±∣)−1,an±,∓=βkngn±,∓=βkn(1+γνoddkn∣un+2±∣+∣un+1∓∣+∣un±∣)−1.(15)\begin{aligned} a_n^{\pm,\mp} &= k_n g_n^{\pm,\mp} = k_n\left(1+\frac{\nu_{\mathrm{odd}}k_n}{\lvert u_{n+2}^{\pm}\rvert+\lvert u_{n+1}^{\mp}\rvert+\lvert u_n^{\pm}\rvert}\right)^{-1},\\ a_n^{\pm,\mp} &= \beta k_n g_n^{\pm,\mp} = \beta k_n\left(1+\frac{\gamma\nu_{\mathrm{odd}}k_n}{\lvert u_{n+2}^{\pm}\rvert+\lvert u_{n+1}^{\mp}\rvert+\lvert u_n^{\pm}\rvert}\right)^{-1}. \tag*{(15)} \end{aligned}

are the wave-affected transfer coefficients. bnb_n and cnc_n are determined from the conservation laws of the energy E=∑n∣un+∣2+∑n∣un−∣2E=\sum_n\lvert u_n^+\rvert^2+\sum_n\lvert u_n^-\rvert^2 and the helicity H=∑nkn∣un+∣2−∑nkn∣un−∣2H=\sum_n k_n\lvert u_n^+\rvert^2-\sum_n k_n\lvert u_n^-\rvert^2 [11], which leads to

bn±,∓=−(1−1λ)an−1∓,∓,cn±,∓=−1λan−2±,∓,bn±,∓=−(1+1λ)an−1±,∓,cn±,∓=1λan−2±,∓.(16)\begin{aligned} b_n^{\pm,\mp} &= -\left(1-\frac{1}{\lambda}\right)a_{n-1}^{\mp,\mp},\\ c_n^{\pm,\mp} &= -\frac{1}{\lambda}a_{n-2}^{\pm,\mp},\\ b_n^{\pm,\mp} &= -\left(1+\frac{1}{\lambda}\right)a_{n-1}^{\pm,\mp},\\ c_n^{\pm,\mp} &= \frac{1}{\lambda}a_{n-2}^{\pm,\mp}. \tag*{(16)} \end{aligned}

The energy balance equation reads [6, 12]

∂tEn=Πn−1−Πn−νkn2En(17)\partial_t E_n = \Pi_{n-1}-\Pi_n-\nu k_n^2 E_n \tag*{(17)}

Πn\Pi_n is the total energy flux

Πn=Πn⇄+Πn⇄,(18)\Pi_n=\Pi_n^{\rightleftarrows}+\Pi_n^{\rightleftarrows}, \tag*{(18)}

where

Πn⇄=Im⁡(an+,†un+2+∗un+1+un++an−,†un+2−∗un+1−un−−cn−,†un+1+∗un−un−1−−cn+,†un+1−∗un+un−1+),Πn⇄=Im⁡(an+,†un+2+∗un+1+un++an−,†un+2−∗un+1−un−−cn−,†un+1−∗un−un−1−−cn+,†un+1+∗un+un−1+).(19)\begin{aligned} \Pi_n^{\rightleftarrows} &= \operatorname{Im}\left(a_{n}^{+,\dagger}u_{n+2}^{+*}u_{n+1}^{+}u_n^{+}+a_n^{-,\dagger}u_{n+2}^{-*}u_{n+1}^{-}u_n^{-}-c_n^{-,\dagger}u_{n+1}^{+*}u_n^{-}u_{n-1}^{-}-c_n^{+,\dagger}u_{n+1}^{-*}u_n^{+}u_{n-1}^{+}\right),\\ \Pi_n^{\rightleftarrows} &= \operatorname{Im}\left(a_n^{+,\dagger}u_{n+2}^{+*}u_{n+1}^{+}u_n^{+}+a_n^{-,\dagger}u_{n+2}^{-*}u_{n+1}^{-}u_n^{-}-c_n^{-,\dagger}u_{n+1}^{-*}u_n^{-}u_{n-1}^{-}-c_n^{+,\dagger}u_{n+1}^{+*}u_n^{+}u_{n-1}^{+}\right). \tag*{(19)} \end{aligned}

are the heterochiral and homochiral components of the flux, respectively. In the main text we use a velocity-based structure function Sp(kn)=⟨(∣un−1+∣∣un+∣∣un+1+∣)p/3⟩S_p(k_n)=\langle(\lvert u_{n-1}^{+}\rvert\lvert u_n^{+}\rvert\lvert u_{n+1}^{+}\rvert)^{p/3}\rangle. The definition reduces the 3-shell oscillation that is observed in shell models [13]. A flux-based structure function SpΠ(kn)=⟨∣Πn/kn∣p/3⟩S_p^{\Pi}(k_n)=\langle\lvert\Pi_n/k_n\rvert^{p/3}\rangle is also used in § 10 [13, 14]. In the main text we mainly focus on the kurtosis K=S4/S22K=S_4/S_2^2. We also explore higher order ‘kurtosis’ K8=S8/S42K_8=S_8/S_4^2, which shows that intermittency is suppressed for structure functions up to p=8p=8 in both DNS and the shell model (Figure S1).

Image not extracted

Figure S1. K8=S8/S42K_8 = S_8/S_4^2 in DNS (a) and for the shell model (b).

We conduct numerical simulations of the two-channel helical shell models with using the 4th order exponential-Runge-Kutta scheme. See Table II for parameters.

ϵ\epsilonν\nuνodd\nu_{\mathrm{odd}}β\betaγ\gammaλ\lambdak0k_0NNTTΔt\Delta t
I1×10−51 \times10^{-5}1×10−91 \times10^{-9}[0.26−66]×10−6[0.26 - 66] \times10^{-6}0.30.30.20.2221/81/830302.5×1052.5 \times10^55×10−45 \times10^{-4}
II1.4×10−51.4 \times10^{-5}9.4×10−69.4 \times10^{-6}[0.3−2.4]×10−3[0.3 - 2.4] \times10^{-3}0.30.30.20.21.6181.6180.7080.70830302.5×1052.5 \times10^55×10−35 \times10^{-3}

Table II. Parameters that are used for the helical shell model in the main text (I) and in Section II D (II). Listed are the energy injection rate ϵ\epsilon, normal shear viscosity ν\nu, odd viscosity νodd\nu_{\mathrm{odd}}, homochiral parameters β\beta and γ\gamma, wave number unit k0k_0, odd viscosity wavenumber kodd=ϵ1/4νodd−3/4k_{\mathrm{odd}} = \epsilon^{1/4}\nu_{\mathrm{odd}}^{-3/4}, total shell number NN, total simulation time TT, time-step Δt\Delta t. Note that parameters (II) of the shell model mimic parameters (I) of DNS.

Dissipation Scale and Extended Inertial Range

As discussed in Ref. [3] and the main text, odd viscosity affects the dissipation scale kck_c with kc∼ϵ1/4ν−1/2νodd−1/4k_c\sim\epsilon^{1/4}\nu^{-1/2}\nu_{\mathrm{odd}}^{-1/4}. This change of the dissipation scale is due to the change of the energy spectrum caused by odd viscosity: odd viscosity arrest direct cascade and changes the spectrum from E(k)∼k−5/3E(k) \sim k^{-5/3} to E(k)∼k−1E(k) \sim k^{-1}, which enhances the dissipation rate at given kk simultaneously. Therefore, when odd viscosity is present the dissipation dominates at larger lengthscales kck_c. In Figure S2,Figure S2) we plot the rescaled energy spectrum E(k)/E0E(k)/E_0 in both DNS and the shell model (E0E_0 is the spectrum for νodd=0\nu_{\mathrm{odd}} = 0). We find that E(k)/E0E(k)/E_0 increases with kk for kodd<k<kck_{\mathrm{odd}} < k < k_c because of the arrested cascade. For k>kck > k_c, E(k)/E0E(k)/E_0 decreases with kk due to the dissipation. This effectively leads to a wavelength selection phenomenon, which is also known as pattern formation [3].

Image not extracted

Figure S2. Wavelength selection in odd turbulence. (a) Rescaled energy spectrum E/E0E/E_0 (E0E_0 is the spectrum for νodd=0\nu_{\mathrm{odd}} = 0) vs wavenumber in DNS for νodd=3×10−4,6×10−4,1.2×10−3,2.4×10−3\nu_{\mathrm{odd}} = 3 \times10^{-4}, 6 \times10^{-4}, 1.2 \times10^{-3}, 2.4 \times10^{-3}. (b) Rescaled energy spectrum vs wavenumber in the shell model for νodd=2.6×10−7,1.0×10−6,4.1×10−6,1.6×10−5,6.6×10−5\nu_{\mathrm{odd}} = 2.6 \times10^{-7}, 1.0 \times10^{-6}, 4.1 \times10^{-6}, 1.6 \times10^{-5}, 6.6 \times10^{-5}. (c) Rescaled energy spectrum estimated from the bootstrap method for νodd=16,256,4096,65536\nu_{\mathrm{odd}} = 16, 256, 4096, 65536, ν=1\nu= 1 and ϵ=1\epsilon= 1. In (a-c), stars indicate condensation scale kc∼νodd−1/4k_c \sim\nu_{\mathrm{odd}}^{-1/4}.

Here we provide a simple bootstrap method which reproduces the pattern formation in odd turbulence. Similar to the shell model, we divide the system into logarithmically distributed shells kn=k02nk_n = k_0 2^n. The velocity fluctuations in each shell is unu_n with shell energy En=un2E_n = u_n^2. Assuming kn>koddk_n > k_{\mathrm{odd}}, the wave-affected energy transfer rate from shell nn to n+1n + 1 is ϵn∼νodd−1un4=νodd−1En2\epsilon_n \sim\nu_{\mathrm{odd}}^{-1}u_n^4 = \nu_{\mathrm{odd}}^{-1}E_n^2 (see § 6). In the steady state, the energy conservation at shell n+1n + 1 requires ϵn=ϵn+1+νkn+12En+1\epsilon_n = \epsilon_{n+1} + \nu k_{n+1}^2 E_{n+1}, i.e., the energy injected in the shell equals to the energy ejected from the shell plus dissipation. Because this equation holds for all shells in the steady state, an estimated energy spectrum can be calculated from an iteration of the energy conservation of all shells, assuming ϵ0=ϵ\epsilon_0 = \epsilon. The results are shown in Figure S2. We indeed find pattern formation at kck_c, and the dependence of kck_c on νodd\nu_{\mathrm{odd}} is consistent with our expectation.

Because kck_c decreases for increasing νodd\nu_{\mathrm{odd}}, the inertial range also shrinks, and the statistics of the kurtosis is affected more by the dissipation when odd viscosity is present. To avoid this effect, we conduct simulations with extended inertial range for both DNS and the shell model. This is achieved by reducing the viscosity to ν~∼ν3/2νodd−1/2\tilde{\nu} \sim\nu^{3/2}\nu_{\mathrm{odd}}^{-1/2} for large νodd\nu_{\mathrm{odd}}. This approach ensures that the inertial range remains approximately same for different νodd\nu_{\mathrm{odd}} values, see Figure S3,Figure S3). The suppression of intermittency is again observed in both DNS and the shell model (Figure S3,Figure S3,Figure S3,Figure S3)). Interestingly, in DNS we observe a weak non-monotonicity of the kurtosis, in alignment with the shell model result and the prediction of the modified Parisi-Frisch theory.

Image not extracted

Figure S3. Simulation results with extended inertial range. The value of ν\nu is reduced when odd viscosity is present, such that the dissipation scale is unchanged. (a) Energy flux in DNS for νodd=0,2.0×10−2\nu_{\mathrm{odd}} = 0, 2.0 \times10^{-2}. (b) Probability distribution of xx-direction vorticity in DNS (renormalized by the standard deviation), for same νodd\nu_{\mathrm{odd}} values as in (a). Gaussian distribution is shown in dashed line for comparison. (c) Kurtosis KK in DNS, for same νodd\nu_{\mathrm{odd}} as in (a). (d) Energy flux in the shell model for νodd=0,2.6×10−7,1.0×10−6,4.1×10−6,1.6×10−5,6.6×10−5\nu_{\mathrm{odd}} = 0, 2.6 \times10^{-7}, 1.0 \times10^{-6}, 4.1 \times10^{-6}, 1.6 \times10^{-5}, 6.6 \times10^{-5}. (e) Probability distribution of Re⁡(u15+)\operatorname{Re}(u_{15}^{+}) in the shell model (renormalized by the standard deviation), for same νodd\nu_{\mathrm{odd}} values as in (d). Gaussian distribution is shown in dashed line for comparison. (f) Kurtosis KK in the shell model, for same νodd\nu_{\mathrm{odd}} as in (d).

Results for smaller scale separation

In the main text, for both the modified Paris-Frisch theory and the shell model we use much larger scale separation than what is accessible in DNS. Such a large scale separation allows us to separate the results into clearly-defined regimes: for kin≪k≪koddk_{\mathrm{in}} \ll k \ll k_{\mathrm{odd}}, there is a direct cascade of ordinary turbulence; for kodd≪k≪kck_{\mathrm{odd}} \ll k \ll k_c, there is a

direct cascade of odd-wave-affected turbulence. Such a distinction between the two regimes requires a sufficiently large separation between the injection scale kink_{\mathrm{in}}, the cross-over scale koddk_{\mathrm{odd}} and the dissipation scale kck_{\mathrm{c}}, and the scale separation in DNS is limited by computational expense.

In this section we provide results for the modified Parisi-Frisch theory and the shell model, using similar scale separation as that in DNS. The results exhibit similar qualitative features as that observed in DNS, validating the two theoretical approaches.

Specifically, for the modified Parisi-Frisch theory we ensure the same odd viscosity scale kodd∈[27−6]k_{\mathrm{odd}} \in[27-6] as that in DNS. In the modified Parisi-Frisch theory, we define kodd=1/rodd(h=1/3)k_{\mathrm{odd}} = 1/r_{\mathrm{odd}(h=1/3)}, where rodd(h)=νodd1/(1+h)r_{\mathrm{odd}}(h) = \nu_{\mathrm{odd}}^{1/(1+h)}. This corresponds to νodd∈[1.2−9.9]×10−2\nu_{\mathrm{odd}} \in[1.2-9.9] \times10^{-2}. As shown in Figure S4, the modified Parisi-Frisch theory reproduces the suppression of kurtosis by odd viscosity. Note that for rr close to 1 we observe an unphysical discontinuity. This is because the original Parisi-Frisch theory is designed only for the asymptotic behavior of the intermittency at small scales, and is ill-defined at large scales.

Comparison between DNS, modified Parisi-Frisch and shell model results

Figure S4. Comparison between DNS, modified Parisi-Frisch and shell model results, for same scale separation as that in DNS. (a) and (c) Energy spectrum and kurtosis in DNS for νodd=0\nu_{\mathrm{odd}}=0, 3×10−43 \times10^{-4}, 6×10−46 \times10^{-4}, 1.2×10−31.2 \times10^{-3}, 2.4×10−32.4 \times10^{-3}, same as that in Figure 3 of the main text. (d) Kurtosis in the modified Parisi-Frisch theory for νodd=0\nu_{\mathrm{odd}}=0, 1.2×10−21.2 \times10^{-2}, 2.5×10−22.5 \times10^{-2}, 4.9×10−24.9 \times10^{-2}, 9.9×10−29.9 \times10^{-2}. (b) and (d) Energy spectrum and rescaled kurtosis in the shell model for νodd=0\nu_{\mathrm{odd}}=0, 3×10−43 \times10^{-4}, 6×10−46 \times10^{-4}, 1.2×10−31.2 \times10^{-3}, 2.4×10−32.4 \times10^{-3}. In (e) kurtosis is rescaled with K~=2.4K\widetilde{K}=2.4K, which ensures that K~≈3\widetilde{K}\approx3 at the forcing scale.

For the shell model we use exactly the same ϵ\epsilon, ν\nu and νodd\nu_{\mathrm{odd}} values as that used in DNS. The energy spectrum of the shell model reproduces the arrest of direct cascade, see Figure S4. The suppression of kurtosis by odd viscosity is also observed in Figure S4 for scale above the dissipation scale. For scale below the dissipation scale, we find anomalous increase of kurtosis, which is a known issue for most shell models [15].

Intermittency suppressed by breakdown of multiple scale invariances

As mentioned in the main text, intermittency is suppressed in odd-wave-affected turbulence, because the multiple scale invariances of the Navier-Stokes equation are broken by odd viscosity. In this section we detail how the multiple scale invariances are broken by the parity-breaking waves produced by odd viscosity.

We start with the inertial-range Navier-Stokes equation

∂tu+u⋅∇u=−∇P+νodde^z×Δu,(20)\partial_t u + u \cdot\nabla u = -\nabla P + \nu_{\mathrm{odd}}\hat{e}_z \times\Delta u, \tag*{(20)}

In the absence of νodd\nu_{\mathrm{odd}}, the inertial Navier-Stokes equation is invariant under global scale transformation [16]

x→λx,u→λhu,t→λ1−ht,(21)x \to\lambda x,\quad u \to\lambda^{h}u,\quad t \to\lambda^{1-h}t, \tag*{(21)}

for arbitrary exponent hh. We define this property as the multiple global scale invariances, which implies that singularities (flow with local self-similar structure) δru∼rh\delta_r u \sim r^h with arbitrary hh can develop in turbulent flow. In Parisi-Frisch multifractal formalism, the turbulent flow contains singularities with a range of exponents hh that are distributed in different locations in the 3D space, which further leads to intermittency [16, 17].

For non-zero νodd\nu_{\mathrm{odd}}, because of the odd term, the global scale invariance only holds for h=−1h=-1. All other global scale invariances are broken by the linear odd term. This conclusion can be generalized to any wave-turbulence system described in Ref. [18], where a wave-generating linear term breaks multiple global scale invariances. However, the breaking of multiple global scale invariances does not necessarily suppress intermittency. As an example, we consider a flow described by Eq. (20) and artificially block the energy transfer between all modes with kz≠0k_z \ne0. This enforces a 2D flow with kz=0k_z=0 on which odd viscosity has no effect at all, and 2D singularities with arbitrary hh are still allowed by scale symmetry (although in reality 2D turbulence is non-intermittent for a different reason [19, 20], see discussion at the end of the section). In other words, while the global scale invariance of Eq. (20) only holds for h=−1h=-1, its 2D part has partial scale invariances for all hh.

This example shows that what matters is not the global scale invariance of the complete Navier-Stokes equation, but is the partial scale invariance of the energy-transferring part of the equation. The non-energy-transferring part of the Navier-Stokes equation does not contribute to the turbulent cascade, hence it does not affect the intermittency.

To identify the energy-transferring part of Eq. (20), we perform helical decomposition,

u(t,x)=∑k∑s=±us(t,k)js(k)eiωs(k)t+ik⋅x,(22)u(t,x)=\sum_k\sum_{s=\pm}u_s(t,k)j^s(k)e^{i\omega_s(k)t+i k\cdot x}, \tag*{(22)}

where j±(k)=e^(k)×(k/∣k∣)±ie^(k)j^\pm(k)=\hat{e}(k)\times(k/|k|)\pm i\hat{e}(k) with e^(k)=e^z×k/∣e^z×k∣\hat{e}(k)=\hat{e}_z\times k/|\hat{e}_z\times k| and frequency

ω±(k)=±νoddkz∣k∣.(23)\omega_\pm(k)=\pm\nu_{\mathrm{odd}} k_z|k|. \tag*{(23)}

±\pm refers to the definite helicity of each helical mode. The explicit dependence of the frequency on the helicity is a result of the broken parity symmetry in odd-viscous fluids. We rewrite Eq. (20) (in the inertial range) as

∂tusk=∑k+p+q=0sp,sq=±Ck∣p,qeiωˉ(k,p,q)tusp∗usq∗.(24)\partial_t u_{s_k}=\sum_{\substack{\boldsymbol{k}+\boldsymbol{p}+\boldsymbol{q}=0\\s_p,s_q=\pm}} C_{k|p,q}e^{i\bar{\omega}(\boldsymbol{k},\boldsymbol{p},\boldsymbol{q})t}u_{s_p}^{*}u_{s_q}^{*}. \tag*{(24)}

Here, a triad with wave vectors (k,p,q)(\boldsymbol{k},\boldsymbol{p},\boldsymbol{q}) and definite helicities (sk,sp,sq)(s_k,s_p,s_q) is a basic unit for energy transfer.

Ck∣p,q=−14(spp−sqq)[(jsp(p)×jsq(q))⋅jsk(k)]∗(25)C_{k|p,q}=-\frac{1}{4}(s_p p-s_q q)\left[(\boldsymbol{j}^{s_p}(\boldsymbol{p})\times\boldsymbol{j}^{s_q}(\boldsymbol{q}))\cdot\boldsymbol{j}^{s_k}(\boldsymbol{k})\right]^{*} \tag*{(25)}

is a coefficient that has the dimension of wave number. ωˉ(k,p,q)=ωsk(k)+ωsp(p)+ωsq(q)\bar{\omega}(\boldsymbol{k},\boldsymbol{p},\boldsymbol{q})=\omega_{s_k}(\boldsymbol{k})+\omega_{s_p}(\boldsymbol{p})+\omega_{s_q}(\boldsymbol{q}) is an important frequency of the triad (k,p,q)(\boldsymbol{k},\boldsymbol{p},\boldsymbol{q}). As we discussed in the main text, energy transfer is effectively blocked when ωˉτeddy≫1\bar{\omega}\tau_{\mathrm{eddy}}\gg1, because the fast oscillation of eiωˉte^{i\bar{\omega}t} cancels the average energy transfer. Therefore, we may divide all triads into quasi-resonant triads (ωˉτeddy≤1)(\bar{\omega}\tau_{\mathrm{eddy}}\le1) and quasi-non-resonant triads (ωˉτeddy>1)(\bar{\omega}\tau_{\mathrm{eddy}}>1). The energy transfer is approximately unaffected by odd viscosity in quasi-resonant triads, and is blocked in quasi-non-resonant triads.

The energy-transferring part of Eq. (20) is then a decimated Navier-Stokes equation in which only quasi-resonant triads are preserved:

∂tusk=∑k+p+q=0sp,sq=±ωˉτeddy≤1Ck∣p,qusp∗usq∗.(26)\partial_t u_{s_k}=\sum_{\substack{\boldsymbol{k}+\boldsymbol{p}+\boldsymbol{q}=0\\s_p,s_q=\pm\\\bar{\omega}\tau_{\mathrm{eddy}}\le1}}C_{k|p,q}u_{s_p}^{*}u_{s_q}^{*}. \tag*{(26)}

While Eq. (26) has a similar form as that of the original Navier-Stokes equation, a different nonlinear dependence on uu emerges from the number of resonant triads, because τeddy∼1/(ku)\tau_{\mathrm{eddy}}\sim1/(ku).

As discussed in the main text, we consider quasi-local triads (k=∣k∣≈∣p∣≈∣q∣)(k=|\boldsymbol{k}|\approx|\boldsymbol{p}|\approx|\boldsymbol{q}|) which dominate the energy cascade. Letting uku_k be the average velocity fluctuations at wave number kk, Eq. (26) is (statistically) simplified to

∂tuk∼gkkuk2,(27)\partial_t u_k\sim g_kku_k^2, \tag*{(27)}

where gkg_k is the fraction of quasi-resonant triads. For homochiral triads which transfer energy inversely (sp=sk=sq)(s_p=s_k=s_q), ωˉ/νodd=skkz∣k∣+sppz∣p∣+sqqz∣q∣≈sk∣k∣(kz+pz+qz)=0\bar{\omega}/\nu_{\mathrm{odd}}=s_k k_z|\boldsymbol{k}|+s_p p_z|\boldsymbol{p}|+s_q q_z|\boldsymbol{q}|\approx s_k|\boldsymbol{k}|(k_z+p_z+q_z)=0 (because k+q+p=0\boldsymbol{k}+\boldsymbol{q}+\boldsymbol{p}=0), leading to gk≈1g_k\approx1. Substituting gkg_k in Eq. (27), we have the following scale invariances of the homochiral channel:

k→λ−1k,uk→λhuk,t→λ1−ht,(28)k\to\lambda^{-1}k,\quad u_k\to\lambda^h u_k,\quad t\to\lambda^{1-h}t, \tag*{(28)}

for arbitrary hh. This is identical to Eq. (21).

For heterochiral triads which carries energy forward (sp=−sk=sq)(s_p=-s_k=s_q), ωˉ/νodd=skkz∣k∣+sppz∣p∣+sqqz∣q∣≈2sk∣k∣kz\bar{\omega}/\nu_{\mathrm{odd}}=s_k k_z|\boldsymbol{k}|+s_p p_z|\boldsymbol{p}|+s_q q_z|\boldsymbol{q}|\approx2s_k|\boldsymbol{k}|k_z. For given kk, ωˉ\bar{\omega} is uniformly distributed in [−2k2νodd,2k2νodd][-2k^2\nu_{\mathrm{odd}},2k^2\nu_{\mathrm{odd}}]. Thus, gk=1g_k=1 for νodd≤uk/(2k)\nu_{\mathrm{odd}}\le u_k/(2k) and gk=uk/(2kνodd)g_k=u_k/(2k\nu_{\mathrm{odd}}) for νodd>uk/(2k)\nu_{\mathrm{odd}}>u_k/(2k). In the wave-affected regime, substituting gk∼uk/kg_k\sim u_k/k in Eq. (27), we have the following scale invariances of the heterochiral channel:

k→λ−1k,uk→λhuk,t→λ−2ht,(29)k\to\lambda^{-1}k,\quad u_k\to\lambda^h u_k,\quad t\to\lambda^{-2h}t, \tag*{(29)}

for arbitrary hh. Note that the scale transformations applied on tt are different in Eqs. (28),(29). Because energy is transferred through homochiral and heterochiral channels simultaneously, a singularity must satisfies both Eqs. (28),(29). In that case, the only scaling exponent consistent with both channels is h=−1h=-1, and the multiple partial scale invariances are broken. In this case, the flow is statistical self-similar, i.e., δru∼rh\delta_r u\sim r^h only holds for a single hh. Because the energy transfer is dominated by the forward flux of heterochiral triads, we have ϵ∼gkkuk3∼νodd−1uk4\epsilon\sim g_kku_k^3\sim\nu_{\mathrm{odd}}^{-1}u_k^4. In this case, uk∼νodd1/4k0u_k\sim\nu_{\mathrm{odd}}^{1/4}k^0 and for consistency we have δru∼νodd1/4r0\delta_r u\sim\nu_{\mathrm{odd}}^{1/4}r^0. This conclusion is used in the modified Parisi-Frisch theory in the main text (Eq. (2)).

The inconsistent scale invariances of homochiral and heterochiral triads are due to the asymmetric effects of odd viscosity on them, i.e., ωˉhomo≠ωˉhetero\bar{\omega}_{\mathrm{homo}}\ne\bar{\omega}_{\mathrm{hetero}}, which originates from the broken-parity symmetry of odd fluids. Indeed, the mechanism discussed above is only valid when the wave frequency explicitly depends on the definite helicity, which requires broken parity. This mechanism can be generalized to any generalized odd waves with frequency ω±(k)=±coddkz/∣k∣α−1\omega_{\pm}(k)=\pm c_{\mathrm{odd}}k_z/|\boldsymbol{k}|^{\alpha-1}, as we discuss in § 7, or any parity-breaking waves. Their parity-preserving counterparts, e.g., waves with frequency ω±(k)=c±kz∣k∣α−1\omega_{\pm}(\boldsymbol{k})=c_{\pm}k_z|\boldsymbol{k}|^{\alpha-1} that does not depend on helicity, do not break the multiple partial scale invariances of the Navier-Stokes equation. This is supported by our shell model results, see § 10. Intermittency has also been experimentally reported in parity-preserving wave turbulence, e.g., gravity capillary wave turbulence [21, 22].

Another consequence of the asymmetric effect of odd viscosity on homochiral and heterochiral channels is the enhanced flux loop, as shown in the main text. Because the homochiral and heterochiral channels transfer energy in different directions, a weak inverse flux in the homochiral channel exists in the direct cascade of ordinary turbulence [23]. The inverse flux of the homochiral channel, together with the forward flux of the heterochiral channel, forms a flux loop, which is amplified by odd viscosity for k>koddk > k_{\mathrm{odd}}. More strikingly, when forcing at small scale kin>koddk_{\mathrm{in}} > k_{\mathrm{odd}}, a flux loop develops at large scale in the absence of a net energy flux [3]. This is also observed in the shell model (Figure S5).

Image not extracted

Figure S5. Flux loop for small-scale forcing. (a) Energy flux in DNS for νodd=2.4×10−3\nu_{\mathrm{odd}} = 2.4 \times10^{-3}. (b) Energy flux in the shell model for νodd=6.6×10−5\nu_{\mathrm{odd}} = 6.6 \times10^{-5}. Forcing is applied on n=13n = 13.

Importantly, the breakdown of multiple scale invariances requires zero or small net helicity injection, which ensures that energy is equally distributed on waves with both definite helicities. When large helicity is injected, the flow can be dominated by waves with the same definite helicity as the injected one. In this case, the nonlinear transfer is dominated by the homochiral channel only, and the asymmetry between the two channels has small effect. Indeed, intermittency has been reported in rotating turbulence (parity is broken by the rotation direction) when large helicity is injected [24].

Another turbulent system that is known to be non-intermittent is 2D turbulence. Although this non-intermittency in 2D has not been fully understood, it must be governed by a different mechanism from what is discussed above. The reason is that the suppression of intermittency in 3D odd turbulence is due to the breakdown of scale invariances. 2D turbulence, on the other hand, has exactly the same scale invariances as that in 3D ordinary turbulence. Hence, intermittency is allowed in 2D from a perspective of scale invariance. It is conjectured that the absence of intermittency in 2D turbulence is related to the energy being transferred from fast degrees of freedom to slower degrees of freedom [25].

Generalized Odd Viscosity

In the main text we discuss the effect of odd viscosity on turbulent fluids, which produces linear waves with frequencies

ω±(k)=±νoddkz∣k∣.(30)\omega_{\pm}(\boldsymbol{k}) = \pm\nu_{\mathrm{odd}} k_z \lvert\boldsymbol{k} \rvert. \tag*{(30)}

In this section we generalize the discussion to generalized odd viscosity with wave frequencies

ω±(k)=±coddkz∣k∣α−1(31)\omega_{\pm}(\boldsymbol{k}) = \pm c_{\mathrm{odd}} k_z \lvert\boldsymbol{k} \rvert^{\alpha-1} \tag*{(31)}

where coddc_{\mathrm{odd}} is a generalized odd viscosity. α\alpha can be any real number. The case for odd viscosity discussed in the main text corresponds to α=2\alpha= 2. When α=0\alpha= 0, one can show that Eq. (31) is equivalent to the Navier-Stokes equation for rotating turbulence with rotating frequency Ω=codd/2\Omega= c_{\mathrm{odd}}/2 [26]. Importantly, all generalized odd waves are parity-breaking waves, since the frequency explicitly depends on the chirality.

We can repeat the analysis of the resonance condition for homochiral and heterochiral triads, as we do in the main text and § 6 for the α=2\alpha= 2 case. For heterochiral triads with locality (k=∣k∣≈p≈∣p∣≈q≈∣q∣)(k = \lvert\boldsymbol{k} \rvert\approx p \approx\lvert\boldsymbol{p} \rvert\approx q \approx\lvert\boldsymbol{q} \rvert) we find the fraction of resonant triads gk=1g_k = 1 for codd≤uk/(2kα−1)c_{\mathrm{odd}} \leq u_k/(2k^{\alpha-1}) and gk=uk/(2kα−1codd)g_k = u_k/(2k^{\alpha-1}c_{\mathrm{odd}}) for codd>uk/(2kα−1)c_{\mathrm{odd}} > u_k/(2k^{\alpha-1}). This modified the energy transfer rate to ϵ∼codd−1k2−2αuk4\epsilon\sim c_{\mathrm{odd}}^{-1} k^{2-2\alpha}u_k^4, which supports a steady-state, wave-affected solution uk∼ϵ1/4codd1/4kα/4−1/2u_k \sim\epsilon^{1/4}c_{\mathrm{odd}}^{1/4}k^{\alpha/4-1/2}. The scaling of the wave-affected solution is consistent with that obtained from the weak-turbulence theories [18, 26, 27]. The ’generalized odd’ scale kodd=ϵ1/(3α−2)codd3/(2−3α)k_{\mathrm{odd}} = \epsilon^{1/(3\alpha-2)}c_{\mathrm{odd}}^{3/(2-3\alpha)} is found by equating the wave-affected solution and the Kolmogorov solution. The dissipation scale kck_c can be estimated from ϵ∼νkc2ukc2\epsilon\sim\nu k_c^2u_{k_c}^2, which leads to kc∼ϵ1/(α+2)codd−1/(α+2)ν−2/(α+2)k_c \sim\epsilon^{1/(\alpha+2)}c_{\mathrm{odd}}^{-1/(\alpha+2)}\nu^{-2/(\alpha+2)}.

For homochiral triads with locality we again find gk=1g_k = 1, i.e., all local triads are resonant. Here, note that homochiral triads with exact locality do not transfer energy, because Ck∣p,k=0C_{k|p,k} = 0 when k=p=qk = p = q. Therefore, homochiral triads can only transfer energy through weakly non-local triads, which are also weakly affected by the resonance condition when α≠1\alpha\ne1. For α=1\alpha= 1, on the other hand, we find that the resonance condition is satisfied by all homochiral triads regardless of the locality. A similar situation also happens in MHD turbulence, in which the frequencies of the Alfvén waves coincide with that of generalized odd waves with α=1\alpha= 1 [27].

For arbitrary α\alpha value, the effects of generalized odd viscosity on homochiral and heterochiral channels are asymmetric. Hence, the multiple partial scale invariances discussed in § 6 are broken for any α\alpha in the wave-affected regime.

Modified Parisi-Frisch

We now extend the modified Parisi-Frisch formalism to generalized odd viscosity. In this case, self-similar solutions with δru∼rh\delta_r u \sim r^h at large scales reduce to the wave-affected solution δru∼codd1/4r1/2−α/4\delta_r u \sim c_{\mathrm{odd}}^{1/4}r^{1/2-\alpha/4} (because uk∼ϵ1/4codd1/4kα/4−1/2u_k \sim\epsilon^{1/4}c_{\mathrm{odd}}^{1/4}k^{\alpha/4-1/2}). The crossover is phenomenologically written as

δru=rh[1+rodd(h)/r]h−1/2+α/4,Ph(r)=rF(h)[1+rodd(h)/r]F(h),(32)\begin{aligned} \delta_r u &= r^h \left[1+r_{\mathrm{odd}}(h)/r\right]^{h-1/2+\alpha/4},\\ P_h(r) &= r^{F(h)}\left[1+r_{\mathrm{odd}}(h)/r\right]^{F(h)}, \tag*{(32)} \end{aligned}

which is same as Eq. (2) of the main text. Here, the crossover lengthscale rodd(h)r_{\mathrm{odd}}(h) is determined by equating two timescales: the eddy turnover time τ=r/δru\tau= r/\delta_r u and the timescale of generalized odd waves τodd=codd−1rα\tau_{\mathrm{odd}} = c_{\mathrm{odd}}^{-1}r^\alpha, which gives rodd(h)=codd1/(h+α−1)r_{\mathrm{odd}}(h) = c_{\mathrm{odd}}^{1/(h+\alpha-1)}. We adopt a commonly-used empirical choice of F(h)F(h): F(h)=2−c1(h−1/9)+c2(h−1/9)ln⁡(h−1/9)F(h) = 2-c_1(h-1/9)+c_2(h-1/9)\ln(h-1/9), with c1=3[[1+ln⁡(ln⁡(3/2))]/ln⁡(3/2)−1]c_1 = 3[[1+\ln(\ln(3/2))]/\ln(3/2)-1] and c2=3/ln⁡(3/2)c_2 = 3/\ln(3/2) [28].

The suppression of intermittency at small lengthscales holds for arbitrary α>2/3\alpha> 2/3, which exhibits the same qualitative features as the results for α=2\alpha= 2 shown in the main text. For α<2/3\alpha< 2/3, the generalized odd viscosity dominates at large lengthscales, e.g., in rotating turbulence, where Eq. (32) becomes invalid.

Two-channel helical shell model

We generalize here the two-channel helical shell model to include generalized odd viscosity. This is achieved by replacing Eq. (15) with

an±,∓=kngn±,∓=kn(1+coddknα−1∣un+2∓∣+∣un+1∓∣+∣un∓∣)−1,(33)a_n^{\pm,\mp}=k_ng_n^{\pm,\mp}=k_n\left(1+\frac{c_{\mathrm{odd}}k_n^{\alpha-1}}{\lvert u_{n+2}^{\mp}\rvert+\lvert u_{n+1}^{\mp}\rvert+\lvert u_n^{\mp}\rvert}\right)^{-1}, \tag*{(33)}
an±,∓=βkngn±,∓=βkn(1+γcoddknα−1∣un+2±∣+∣un+1±∣+∣un±∣)−1.a_n^{\pm,\mp}=\beta k_ng_n^{\pm,\mp}=\beta k_n\left(1+\frac{\gamma c_{\mathrm{odd}}k_n^{\alpha-1}}{\lvert u_{n+2}^{\pm}\rvert+\lvert u_{n+1}^{\pm}\rvert+\lvert u_n^{\pm}\rvert}\right)^{-1}.

The value of γ\gamma controls which channel is affected more by generalized odd viscosity. According to our analysis of resonant triads in the main text, γ<1\gamma< 1 such that odd viscosity has a weaker effect on the homochiral channel. We show here the results for α=1.5\alpha= 1.5, which exhibits the same qualitative features as the α=2\alpha= 2 case in the main text (Figure S6,Figure S6).

Six-panel plots of flux and kurtosis for parity-breaking and parity-preserving waves

Figure S6. Shell model simulations for wave frequencies ω(k)∼coddkz∣k∣α−1\omega(k) \sim c_{\mathrm{odd}} k_z |k|^{\alpha-1} with α=1.5\alpha= 1.5. Inertial range is extended by reducing the normal viscosity, see Sec. II C. (a) Energy flux for parity-breaking waves (γ<1\gamma< 1) and codd=8.4×10−3c_{\mathrm{odd}} = 8.4 \times10^{-3}. (b) Velocity-based kurtosis for parity-breaking waves (γ<1\gamma< 1) and codd=0.8,3×10−6,3.3×10−5,1.3×10−4,5.2×10−4,2.1×10−3,8.4×10−3c_{\mathrm{odd}} = 0.8, 3 \times10^{-6}, 3.3 \times10^{-5}, 1.3 \times10^{-4}, 5.2 \times10^{-4}, 2.1 \times10^{-3}, 8.4 \times10^{-3} (c) Flux-based kurtosis for parity-breaking waves (γ<1\gamma< 1) and same coddc_{\mathrm{odd}} values as in (b). (d) Energy flux for parity-preserving waves (γ=1\gamma= 1) and codd=8.4×10−3c_{\mathrm{odd}} = 8.4 \times10^{-3}. (e) Velocity-based kurtosis for parity-preserving waves (γ=1\gamma= 1) and coddc_{\mathrm{odd}} values same as in (b). (f) Flux-based kurtosis for parity-preserving waves (γ=1\gamma= 1) and same coddc_{\mathrm{odd}} values as in (b).

As shown in Figure 4(a), generalized odd viscosity modifies the scaling behavior of the structure functions for k≫koddk \gg k_{\mathrm{odd}}. This can be quantitatively described by a differential scaling exponent zp(k)=−dlog⁡(Sp)/dlog⁡(k)z_p(k)=-\mathrm{d}\log(S_p)/\mathrm{d}\log(k). We plot z2z_2 and z6z_6 in the shell model for different α\alpha and coddc_{\mathrm{odd}} values, vs rescaled wavenumber k/koddk/k_{\mathrm{odd}}, see Figure S7-Figure S7, e.g. We find that zpz_p changes from the usual multifractal exponents of regular turbulence for k≪koddk \ll k_{\mathrm{odd}} to a plateau for k≫koddk \gg k_{\mathrm{odd}}. The plateau corresponds to a fully wave-affected regime where Sp(kn)∼kn−ζpS_p(k_n) \sim k_n^{-\zeta_p}. We fit the values of ζp\zeta_p using the shaded area, and the results are shown in Figure 4d of the main text. We then fit ζp\zeta_p using the monofractal formula ζp=hαp\zeta_p=h_\alpha p. The numerical values of hαh_\alpha are close to the theoretical approximation ζp=p(1/2−α/4)\zeta_p=p(1/2-\alpha/4) (Figure S7). The theoretical hαh_\alpha is obtained using the resonant fraction, see below Eq. (31).

Differential scaling exponents in the shell model

Figure S7. Differential scaling exponent of S2S_2 (a-c) and S6S_6 (e-g) in the shell model with generalized odd viscosity (inducing generalized odd waves with frequency ω±(k)=±coddkz∣k∣α−1\omega_{\pm}(k) = \pm c_{\mathrm{odd}}k_z|k|^{\alpha-1}). Exponents are plotted vs rescaled wavenumber k/koddk/k_{\mathrm{odd}}, with kodd=ϵ1/(3α−2)codd3/(2−3α)k_{\mathrm{odd}} = \epsilon^{1/(3\alpha-2)}c_{\mathrm{odd}}^{3/(2-3\alpha)}. Results are shown for codd/c0=0,4mc_{\mathrm{odd}}/c_0 = 0,4^m, with c0=(ϵ/2)1/2−α/43α/4−1/2c_0 = (\epsilon/2)^{1/2-\alpha/4}3^{\alpha/4-1/2}. (a, e) α=1.2\alpha= 1.2 and m=2−6m = 2 - 6. (b, f) α=1.5\alpha= 1.5 and m=3−7m = 3 - 7. (c, g) α=2\alpha= 2 and m=4−9m = 4 - 9. (d) Coefficient hα=ζp/ph_\alpha= \zeta_p/p obtained from the shell model. Theoretical approximation hα=1/2−α/4h_\alpha= 1/2 - \alpha/4 is shown for comparison.

Parity-breaking waves vs parity-preserving waves

As discussed in § 6, the multiple scale invariances of the Navier-Stokes equation are broken by the asymmetric effects of (generalized) odd viscosity on homochiral and heterochiral channels. This mechanism only applies to parity-breaking waves which break parity symmetry. For parity-preserving waves, the effects on homochiral and heterochiral channels are symmetric and intermittency may still develop.

This argument can be tested by the two-channel helical shell model with generalized odd viscosity (§ 7 B). The asymmetry between homochiral and heterochiral channels is controlled by the parameter γ\gamma in Eq. (33): when γ=1\gamma= 1 the two channels are affected in a symmetric way by generalized odd viscosity; when γ<1\gamma< 1 the homochiral channel is less affected. Hence, while the shell model with γ<1\gamma< 1 corresponds to parity-breaking waves with ω±(k)=±coddkz∣k∣α−1\omega_{\pm}(k) = \pm c_{\mathrm{odd}} k_z |k|^{\alpha-1}, the shell model with γ=1\gamma= 1 corresponds to parity-preserving waves with, e.g., ω±(k)=coddkz∣k∣α−1\omega_{\pm}(k) = c_{\mathrm{odd}} k_z |k|^{\alpha-1}.

We explore the intermittency for both the parity-breaking and parity-preserving models with α=1.5\alpha= 1.5. As expected, the flux-loop is enhanced by coddc_{\mathrm{odd}} in the parity-breaking model (Figure S6) but is not enhanced in the parity-preserving model (Figure S6). The kurtosis is suppressed by large coddc_{\mathrm{odd}} in the parity-breaking model (Figure S6). In the parity-preserving model, on the other hand, the kurtosis still increases with a relatively slow speed for large coddc_{\mathrm{odd}}, suggesting that the intermittency is not completely suppressed. This difference is more obvious for the flux-based kurtosis, defined by KΠ=S4Π/S2Π2K^{\Pi} = S^{\Pi}_{4}/S^{\Pi2}_{2}, where SpΠ(kn)=⟨∣Πn/kn∣p/3⟩S^{\Pi}_{p}(k_n) = \langle|\Pi_n/k_n|^{p/3} \rangle is the flux-based structure function [13, 14]. We find that while the parity-breaking model is still non-intermittent in the wave-affected regime, the parity-preserving model is strongly intermittent (Figure S6,Figure S6).

A similar comparison is done for odd waves (α=2\alpha= 2) (Figure S8, comparing a,b,c with d,e,f). Surprisingly, we find that the velocity-based kurtosis is suppressed in both the parity-breaking and parity-preserving models (Figure S8,Figure S8). When looking at the flux-based kurtosis we again observe different results: kurtosis does not develop in the parity-breaking model but develops slowly in the parity-preserving model (Figure S8,Figure S8). This suggests that the parity-breaking model is non-intermittent and the parity-preserving model is intermittent in the wave-affected regime. However, the intermittency of the parity-preserving model is not reflected on the velocity-based kurtosis. This is possibly a

Shell model simulations for wave frequencies

Figure S8. Shell model simulations for wave frequencies ω(k)∼νoddkz∣k∣\omega(k) \sim\nu_{\mathrm{odd}} k_z \lvert k\rvert. Inertial range is extended by reducing the normal viscosity, see Sec. II C. (a) Energy flux for parity-breaking waves (γ<1\gamma< 1) and νodd=6.6×10−5\nu_{\mathrm{odd}} = 6.6 \times10^{-5}. (b) Velocity-based kurtosis for parity-breaking waves (γ<1\gamma< 1) and νodd=0,2.6×10−7,1.0×10−6,4.1×10−6,1.6×10−5,6.6×10−5\nu_{\mathrm{odd}} = 0, 2.6 \times10^{-7}, 1.0 \times10^{-6}, 4.1 \times10^{-6}, 1.6 \times10^{-5}, 6.6 \times10^{-5}. (c) Flux-based kurtosis for parity-breaking waves (γ<1\gamma< 1) and same νodd\nu_{\mathrm{odd}} values as in (b). (d) Energy flux for parity-preserving waves (γ=1\gamma= 1) and νodd=6.6×10−5\nu_{\mathrm{odd}} = 6.6 \times10^{-5}. (e) Velocity-based kurtosis for parity-preserving waves (γ=1\gamma= 1) and νodd\nu_{\mathrm{odd}} values same as in (b). (f) Flux-based kurtosis for parity-preserving waves (γ=1\gamma= 1) and same νodd\nu_{\mathrm{odd}} values as in (b).

α\alphaShellTheory
1.20.1750.2
1.50.1040.125
20.0270

coincidence for α=2\alpha= 2: when α=2\alpha= 2, the wave-affected solution En∼ϵ−1/2νodd−1/2kn0E_n \sim\epsilon^{-1/2}\nu_{\mathrm{odd}}^{-1/2}k_n^0 coincides with the equipartition solution (all unSu_{nS} have similar velocity fluctuations). This leads to strong fluctuations in the wave-affected regime which may mask the non-Gaussian fluctuations of unu_n. For the 3D Navier-Stokes equation the equipartition solution holds at E(k)∼k2E(k) \sim k^2, hence a similar coincidence may happen at α=6\alpha= 6 instead of α=2\alpha= 2.

Shell Model with Stochastic Waves

In our two-channel helical shell model, the generalized odd waves in the Navier-Stokes equation are not explicitly included. The effect of generalized odd waves are modelled by the fractions of resonant triads gng_n. A different approach was adopted in prior shell models of rotating and MHD turbulence, which modifies the classic one-channel shell model with explicit waves [29, 30]. This approach is summarized in the following form:

∂tun=Bn+fn−νkn2un+iΩn(t)un,(34)\partial_t u_n = B_n + f_n - \nu k_n^2 u_n + \mathrm{i}\Omega_n(t)u_n , \tag*{(34)}

where

Bn=anun+2un+1∗+bnun+1un−1∗−cnun−1un−2(35)B_n = a_n u_{n+2}u_{n+1}^{*} + b_n u_{n+1}u_{n-1}^{*} - c_n u_{n-1}u_{n-2} \tag*{(35)}

is the nonlinear transfer of Sabra shell model with an=−2bn+1=−2cn+2=kna_n = -2b_{n+1} = -2c_{n+2} = k_n. It has been argued that to reproduce the wave-affected energy spectrum, the wave frequencies Ωn(t)\Omega_n(t) cannot be assumed constant, but has to be stochastic with ⟨Ωn2⟩∼codd2kn2α\langle\Omega_n^2\rangle\sim c_{\mathrm{odd}}^2 k_n^{2\alpha} [29]. The reason for this may be explained by the resonance condition: In the Navier-Stokes equation, generalized odd waves oscillate with frequencies ω±(k)=±coddkz∣k∣α−1=±codd∣k∣αcos⁡(θ)\omega_{\pm}(k) = \pm c_{\mathrm{odd}}k_z|k|^{\alpha-1} = \pm c_{\mathrm{odd}}|k|^\alpha\cos(\theta), where θ\theta is the angle between kk and e^z\hat{e}_z. The 3D distribution of θ\theta effectively induces a wide distribution of the wave frequencies for given ∣k∣|k|, which further guarantees that the resonant condition can be satisfied by some triads. In shell models, however, this 3D structure is absent and the interaction among triads is limited to nearest neighbors. Hence, if constant frequencies Ωn∼coddknα\Omega_n \sim c_{\mathrm{odd}}k_n^\alpha is assumed, one can show that the resonant condition Ω‾nτeddy<1\overline{\Omega}_n\tau_{\mathrm{eddy}} < 1 will never be satisfied, with Ω‾n=Ωn−1+Ωn+Ωn+1\overline{\Omega}_n = \Omega_{n-1} + \Omega_n + \Omega_{n+1}.

The stochasticity of Ωn\Omega_n in the shell model effectively reproduces the wide distribution of wave frequencies in the Navier-Stokes equation. The difference is that in the shell model, different Ωn\Omega_n is distributed in different time period, while in the Navier-Stokes equation, different frequencies are distributed in different triads. This type of shell models is able to reproduce the average energy transfer of the Navier-Stokes equation, however, it would be inappropriate to study intermittency with these models. The reason is that the stochasticity naturally breaks scale invariances at any given time, hence intermittency must be suppressed. This suppression is just caused by an artifact of the stochastic shell model that is absent in the Navier-Stokes equation.

References

  1. [1]M. Fruchart, M. Han, C. Scheibner, and V. Vitelli, “The odd ideal gas: Hall viscosity and thermal conductivity from non-hermitian kinetic theory,” (2022), arXiv:2202.02037.arxiv.org/abs/2202.02037
  2. [2]T. Khain, C. Scheibner, M. Fruchart, and V. Vitelli, Journal of Fluid Mechanics 934, A23 (2022).
  3. [3]X. M. de Wit, M. Fruchart, T. Khain, F. Toschi, and V. Vitelli, Nature 627, 515 (2024).
  4. [4]P. A. Davidson, Turbulence: an introduction for scientists and engineers (Oxford university press, 2015).
  5. [5]A. Alexakis and L. Biferale, Physics Reports 767–769, 1–101 (2018).
  6. [6]M. K. Verma, Energy transfers in fluid flows: multispectral and spectral perspectives (Cambridge University Press, 2019).DOI
  7. [7]G. Dar, M. K. Verma, and V. Eswaran, Physica D: Nonlinear Phenomena 157, 207 (2001), 0109004 [nlin].
  8. [8]A. Alexakis, P. D. Mininni, and A. Pouquet, Physical Review E 72, 046301 (2005).
  9. [9]P. D. Mininni, A. Alexakis, and A. Pouquet, Physical Review E 72, 046302 (2005).
  10. [10]R. Peyret, Spectral Methods for Incompressible Viscous Flow (Springer, 2002).DOI
  11. [11]R. Benzi, L. Biferale, R. Kerr, and E. Trovatore, Physical Review E 53, 3541 (1996).
  12. [12]L. Biferale, Annual Review of Fluid Mechanics 35, 441–468 (2003).
  13. [13]V. S. L’vov, E. Podivilov, A. Pomyalov, I. Procaccia, and D. Vandenbroucq, Physical Review E 58, 1811 (1998).
  14. [14]X. M. de Wit, G. Ortali, A. Corbetta, A. A. Mailbyaev, L. Biferale, and F. Toschi, Physical Review E 109, 055106 (2024).
  15. [15]R. Benzi and L. Biferale, Journal of Statistical Physics 135, 977 (2009).
  16. [16]G. Parisi and U. Frisch, in Proceedings of the International School of Physics “E. Fermi”, edited by M. Ghil, R. Benzi, and G. Parisi (North-Holland, Amsterdam, 1985) pp. 84–87.
  17. [17]U. Frisch, Turbulence: The Legacy of A. N. Kolmogorov (Cambridge University Press, 1995).
  18. [18]V. Zakharov, V. L’vov, and G. Falkovich, Kolmogorov Spectra of Turbulence I: Wave Turbulence (Springer-Verlag, Berlin, Germany, 1992).
  19. [19]G. Paladin and A. Vulpiani, Physics Reports 156, 147 (1987).
  20. [20]G. L. Eyink and K. R. Sreenivasan, Rev. Mod. Phys. 78, 87 (2006).
  21. [21]E. Falcon, S. Fauve, and C. Laroche, Physical Review Letters 98, 154501 (2007).
  22. [22]S. Lukaschuk, S. Nazarenko, S. McLelland, and P. Denissinko, Physical Review Letters 103, 044501 (2009).
  23. [23]E. Deusebio, G. Boffetta, E. Lindborg, and S. Musacchio, Physical Review E 90, 023005 (2014).
  24. [24]P. D. Mininni and A. Pouquet, Physics of Fluids 22, 035106 (2010).
  25. [25]N. Vladimirova, M. Shavit, and G. Falkovich, Physical Review X 11, 021063 (2021).
  26. [26]S. Galtier, Physical Review E 68, 051301 (2003).
  27. [27]S. Galtier, S. Nazarenko, A. C. Newell, and A. Pouquet, Journal of plasma physics 63, 447 (2000).
  28. [28]Z.-S. She and E. Lévêque, Physical Review Letters 72, 336 (1994).
  29. [29]Y. Hattori, R. Rubinstein, and A. Ishizawa, Physical Review E 70, 046311 (2004).
  30. [30]F. Plunian and R. Stepanov, Physical Review E 82, 046311 (2010).

Paper details

vlnc:W_u3ejbp5d
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents